Monday.com's AI Credit Ledger: The Pricing Pivot Nobody Audited

On May 2026, Monday.com retired the cleanest assumption in enterprise software: a seat is a seat. It replaced that assumption with a hybrid pricing model - a basic subscription plus a metered AI credit system. The stock had collapsed by more than half since the start of the year. The announcement produced a 12.6 percent rebound. The market read a new story. I read a new ledger. And ledgers do not lie, only the auditors do. The context is a company in transition. Monday.com spent years defining the 'Work OS' category. Now it calls itself an 'AI Work Platform.' The shift is not cosmetic. The product is moving from a system of record - a place where work is organised - to a system of action - a place where work is executed by AI agents. The company says it has roughly a quarter-million enterprise customers. It integrates Anthropic, OpenAI, and Microsoft models through native one-click connectors, so non-technical team members can build and deploy AI workflows. It also cut 620-630 employees, about 20 percent of its workforce, to 'adapt the company to our new vision.' It expects restructuring charges of $45 million to $55 million. It reaffirmed 19-20 percent revenue growth. The market rewarded all of this with a 12.6 percent bump. I am not impressed by the bump. I am impressed by the obligation that just landed on Monday.com's balance sheet. Let me show you the meter. The pricing block is not complicated. Basic, Standard, and Pro now include 1,000, 2,000, and 3,000 AI credits per month. Overage costs $0.01 to $0.0125 per credit. A monthly credit plan is roughly 25 percent more expensive than an annual one. That 25 percent is not a detail; it is a working-capital transfer. The customer pays upfront, Monday.com holds the float, and the customer hopes the credits get consumed. In DeFi, we would call that a token sale before mainnet. What is one credit? The announcement does not define it. Is it one inference? One tool call? One workflow step? One completed agent run? If the unit is undefined, the meter can be tuned. I have seen this before. In 2017, I spent 40 hours auditing a distribution contract for a Dublin fintech. The community was loud, but the code was leaking. I found an integer overflow in the bonus calculation: at a certain block height, the token distribution would wrap around and expose the remaining balance to a drain. The bounty earned me 2,000 ETH. The lesson was simple: if you cannot audit the unit of settlement, you are not holding a position; you are holding a story. The credit is a settlement unit. When a customer pre-pays for 12,000 credits, Monday.com records cash and a promise to deliver future AI work. That promise is a liability. If the customer only uses 4,000 credits, the remaining 8,000 remain on the balance sheet. The bigger the credit balance, the bigger the obligation. That is not recurring revenue. That is deferred service revenue with a model-cost risk attached. The word 'credit' gives it an asset-like sound, but to Monday.com it is an unmined liability. Now come the unit economics. This is where the market's new narrative starts to crack. A seat has near-zero marginal cost. An AI credit has a real marginal cost: every time an agent calls Claude, GPT-4, or an Azure model, Monday.com pays the model provider. If the model consumption cost is 40-60 percent of the credit price, the gross margin on credit revenue is 40-60 percent. Let me run the numbers. Suppose the average credit price is $0.011. Suppose a workflow consumes $0.005 of external model cost. The contribution margin is $0.006 per credit - roughly 55 percent. Traditional SaaS margins are in the 75-85 percent range. If AI credits become 20 percent of Monday.com's revenue, the blended gross margin drops from roughly 80 percent to 70-75 percent. That is the difference between a premium software company and a middleman. Investors should watch that margin line closely. The story is not 'AI credits create new revenue.' The story is 'AI credits introduce a new cost line, then multiply existing revenue.' If the external model cost falls faster than the credit price, Monday.com makes more money. If the model cost rises - or if new models require more tokens to produce the same result - the margin disappears. The company has no control over Anthropic's pricing, OpenAI's token efficiency, or Microsoft's enterprise discount schedule. It is selling an unhedged spread on third-party infrastructure. I spent two weeks in January 2024 trading the spread between the spot Bitcoin ETF and the Coinbase Premium Index. I built a Python script to track every tick. That trade worked because I could measure the spread in real time. Monday.com's customers cannot measure their spread at all. The credit price is a black box. Let me give you a worked example. A mid-market finance team of 15 people takes a Standard plan with 2,000 included credits. They automate vendor invoice reconciliation. Each approval workflow costs 2 credits per vendor: one for extraction, one for validation. At 600 vendors per month, they burn 1,200 credits, inside the plan. Then they expand to purchase-order matching. That costs 4 credits per order, and they process 600 orders per month. Total burn is 3,600 credits. They buy the remaining 1,600 credits at $0.01, an overage of $16. That feels harmless. Add an AI assistant that summarizes every new contract at 20 credits per contract, 120 contracts per month, and total burn jumps to 6,000 credits. Overage is now 4,000 credits, or $40. Scale to 100,000 vendors and the overage is nearly $1,000. This is a utility bill. It scales linearly with tasks, not with users. The finance team stops asking how many licenses they need and starts asking how much this operation costs. There is also the revenue-quality question. Does an AI credit purchase count as ARR? Under the old seat model, revenue was predictable: one seat, one price, one month. Under the credit model, revenue is consumption-based. If a customer buys credits but never uses them, the cash came in, but the service obligation remains. Recognising that cash as recurring revenue before delivery is like calling a stablecoin reserve 'equity' before the redemption request. Yield without due diligence is just borrowed luck. The next question is usage. AI agents are supposed to increase usage. But the meter cuts both ways. In the old SaaS world, more product usage meant more value and more retention. In the metered AI world, more efficient AI means fewer credits burned. If Monday.com improves its agents so they can do the same task with 20 percent fewer credits, customers save money. Monday.com loses money. This is the AI efficiency paradox: the better the product, the slower the meter. Unless the credit bundle is designed as a minimum contract, the growth in AI credit consumption can go backwards while the product is improving. This is the core reason why I do not treat the stock bounce as a signal. The old metrics - DAU, MAU, NRR, gross margin - were built for a world where a user interaction cost nothing. In the new world, every interaction has a price. The customer has to ask: how many credits is this task worth? Most businesses do not know. They will either overbuy and waste money, or underbuy and lose faith in AI. Both outcomes hurt Monday.com in different ways. Let me go contrarian now. The consensus is that Monday.com is turning itself from a sluggish work-management vendor into a high-growth AI platform. I think the more accurate description is that Monday.com is building a metering company. The AI Work Platform is a router between two communities: enterprise workflows on one side, external model providers on the other. It takes a request, pushes it to a third-party model, charges a credit, and records the result. That is not a moat; that is a toll booth. Toll booths are only useful as long as traffic cannot take another road. The biggest threat is not Asana or ClickUp. It is Microsoft. Monday.com connects to Microsoft and OpenAI in one move. But Microsoft also sells Teams, Project, and Copilot. If Microsoft extends Copilot into an enterprise agent orchestration layer, it can route the same workflows without Monday.com. The connector strategy is a rented moat. Short-term, it gives Monday.com AI capabilities without building a foundation model. Long-term, it makes Monday.com dependent on the same players it competes against. Liquidity is the only truth in a fragmented chain. In this case, the chain is fragmented across OpenAI, Anthropic, and Microsoft. Monday.com is a liquidity aggregator. The moment one of those pools decides to build its own interface, the flow disappears. There is also the data trust problem. Enterprise customers are not stupid. They know that when an AI agent runs on a third-party model, their workflow data may travel outside their tenant. If Monday.com cannot offer a zero-retention agreement or a private deployment option, large customers will only route low-risk tasks into AI agents. Low-risk tasks consume small credits. The same data-security hesitation that limits cloud adoption in regulated industries will cap the entire AI credit revenue line. Data flywheels sound great in investor decks. They require data to actually flow. If the data does not move, the flywheel does not spin. The flywheel is the most overused word in this pivot. Monday.com has a claimed network effect: thousands of teams build work templates, integrations, and now agents. That is real. But the value of that network depends on sharing operational knowledge. Enterprises do not want their operational knowledge shared. They want their AI agents to be private, their prompts to stay inside the tenant, and their workflows to remain inaccessible to competitors. So the most important data for the flywheel - the data that describes how a customer actually runs its business - is exactly the data that cannot be pooled. The public template can be shared. The proprietary adjustment cannot. So the flywheel is weaker than Monday.com's investor narrative implies. Switching costs behave differently in an AI-native world. In the old Monday.com, leaving meant migrating boards and templates. In the new Monday.com, leaving means rebuilding agents, re-routing workflows, and re-training the model layer on a new orchestration runtime. A customer that spent three months building 40 AI agents for procurement, support, and marketing cannot leave without rewriting those agents. That is a real lock-in. But it is also a reason for new customers to hesitate. They know that once they invest in Monday.com's agent runtime, they are locked to its orchestration model. So they will run a proof-of-concept on the least critical workflow first. That proves safety, not scale. And then there is the customer service question. The company cut 20 percent of its workforce in the same quarter that it introduced a radically more complex pricing model. The customer success team used to focus on teaching users how to use a project board. Now it must help customers estimate AI credit burn, design agent workflows, and prove ROI. That is a different skill set. You do not hire those skills by firing the people who had them. The transition can easily open a twelve-to-eighteen-month gap in which new AI features ship but customer onboarding degrades. In a subscription business, that is not a headline risk. It is a renewal risk. There is also a product complexity angle. The company sells the idea that non-technical team members can configure AI agents with one-click connectors. One-click connectors solve the connection problem. They do not solve state management, permissioning, error recovery, or audit logging. The moment an agent fails in production, a non-technical business user cannot reason about why. They will call customer support. Customer support was just cut 20 percent. The citizen developer story is optimistic. The citizen debugger story is where the compliance officer starts asking questions. Behind the credit meter is a serious infrastructure problem. To bill a customer for 1,000 credits, Monday.com needs to know exactly which agent consumed which model, how many tokens were used, which tools were called, and what output was produced. That is a metering and billing layer as complex as a cloud provider's meter. That is not a one-quarter project. If Monday.com skips this investment, customers will see latency, mystery charges, and reconciliation failures. If it invests properly, it becomes a different company: not a software company, but an infrastructure company. The market is still pricing it as a software company. The hardest sale in this model is the first one. You are asking a customer to budget for something they have never measured. A CFO wants a fixed cost. A sales rep wants to sell outcomes. The product only consumes credits in production. So the sales cycle lengthens, the discount pressure increases, and the sales team goes back to school on value selling. The 19-20 percent growth guidance might still be true, but the path to that guidance is not a beautiful AI story. It is a grind through procurement departments that have already been burned by cloud metering bills. I have one more comparison. This is the same pattern I saw during DeFi Summer in 2020. Projects printed a governance token, attached it to a treasury, and hoped that liquidity would validate it. Some had real yield; most were just shuffling TVL. Monday.com's AI credits are not a governance token, but the accounting logic is similar: raise the cash now, promise to deliver value later, and let the market price the future. There is no code audit for a pricing model. There is only the next quarterly filing. What would change my view? I need three disclosures. First, I need Monday.com to report the split between seat revenue and credit revenue. Second, I need a reconciliation of AI credits purchased versus AI credits consumed. Third, I need a margin bridge showing what percentage of credit revenue is paid to external model providers. If those numbers are transparent, I can calculate the true consumption curve and decide whether this is a real business or a pass-through. If those numbers stay hidden, the 12.6 percent bounce is just beta. Beta is the tax you pay for ignorance. There is a fourth disclosure that would make me more constructive: a credit expiry policy. If Monday.com lets unused credits expire, it gets to keep the float without delivering service. That would be a profitable accounting feature, but it would also be a customer-hostile one. If credits expire too quickly, customers will feel cheated; if they never expire, the deferred liability grows forever. The optimal policy is a five-year expiry with a transparent burn-rate dashboard. That is not just a product feature. It is a balance-sheet strategy. The deeper issue is where this business model lands on the spectrum between software and utility. A pure SaaS company sells outcomes through subscriptions. A pure utility company sells units of consumption. Monday.com is trying to sit in between. It wants the recurring revenue of a subscription and the upside of a utility. That combination is rare. It works only when the unit price is transparent, the meter is accurate, and the customer can forecast consumption without a data-science team. Most enterprise customers cannot forecast AI consumption today. They can barely forecast cloud spending. That is why I am not joining the bullish crowd. I know the crowd is using a simple narrative: AI plus software equals growth. I live in the tangled world of spreadsheets, stop-losses, and metered gas. I know that when a product is hard to price, sales cycles get longer, discounting gets deeper, and margin gets worse before it gets better. Monday.com is not immune to that. It is doing a large, painful thing. Layoffs, a new pricing model, a new product identity, a new sales motion, and a new customer success model are all happening in the same twelve months. That is a lot of moving parts for a company that just reaffirmed 19-20 percent revenue growth. I am not saying the story fails. I am saying the story has not been audited. The meter is running. But the meter is not public. In my world, settlement is only valid when both parties can read the same ledger. Monday.com has built a ledger that only it can read. That is not a feature. It is a risk. And risk should be compensated, not celebrated. The algorithm executes, but the human decides. I have spent 18 years watching markets punish people who confuse a label with a settlement. Monday.com is now a token issuer. Its token is called a credit. Its reserve is called an AI agent runtime. Its redemption obligation is called a workflow result. None of that means it will fail. It means I need to audit the meter before I pay for the story. Sanity checks before sanity wins. I will start checking next quarter. Until then, I would rather watch the flow than chase the bounce. Let the meter be transparent. Then, and only then, will I call it an AI Work Platform. Until then, it is an AI Work Gamble.